scrollback — Sunday June 28

The new Loopy skill now walks you through loop design, executes the loops, and surfaces debug reviews afterward. For someone already spending $440/mo on agent tooling this is a direct fit for repeatable workflows instead of one-off prompts. The GitHub skill installs cleanly and separates the library code from the website, which matters if you're wiring it into Claude Code or a custom MCP setup.

The macOS app pairs local speech models with Fluid-1 post-processing that adapts output for Cursor, terminal, Slack, or Mail without leaving the machine. At under 100 ms perceived latency and full offline support it slots straight into existing terminal-heavy workflows. You pick the model routing and keep everything on-device, which aligns with your cost-per-task focus over cloud token spend.